An adaptive directed transfer function approach for detecting dynamic causal interactions.

This paper proposes the application of the directed transfer function (DTF) to a set of time-varying coefficients obtained through the use of a multivariate adaptive autoregressive (MVAAR) model. We define this time-varying measure of causality as the adaptive directed transfer function (ADTF) and compare its ability to discern changes in the causal interaction pattern as compared with the conventional DTF. To accomplish this task, two multivariate models with predefined interaction patterns were created in which the causal interaction between the nodes was altered during the course of the time series. In both models, the ADTF has the capability to discern the dynamic changes in the primary source of the information outflow. The results obtained by using the ADTF were subsequently compared to those calculated through use of the conventional DTF method and the present simulation result suggests that use of the ADTF could provide useful information regarding dynamic causality.